Design Philosophy

Overview

This package provides a collection of statistical functions covering confidence and prediction interval estimation, hypothesis testing, meta-analysis, resampling-based inference, and utilities for machine learning model development and evaluation, with a focus on epidemiological and clinical research applications.

The emphasis is on correctness, transparency, and reproducibility. Where established methods exist in the literature, this package aims to implement them faithfully and expose the assumptions and limitations clearly—both in code and in documentation.

Users are expected to supply pre-processed inputs; the package does not impose a data pipeline.

Design Principles

• Correctness first: Implementations follow published statistical methods with explicit references. Approximations are documented and alternatives provided where appropriate.

• Transparent results: All functions return named result objects. No quantity computed internally is hidden from the caller.

• Minimal preprocessing: Functions assume input data are already in the correct form. Validation is limited to type and shape checks at the boundary; internal transformations are not hidden.

• Composable by design: Result objects are plain Python dataclasses. They can be passed to other functions, serialised, or inspected in a notebook without any framework dependency.